Google DeepMind has introduced a new AI model called Gemini 2.5 Deep Think. This advanced AI can think about several ideas at the same time and then choose the best answer.
The model is available to Google’s most expensive subscribers—the Ultra plan costing $250 per month—and can be accessed through the Gemini app.
Gemini Deep Think is Google’s first multi-agent AI model. This means it uses multiple AI “agents” working in parallel to solve problems. While this needs more computing power than traditional AI, it usually gives better and more accurate results.
Google used this AI to win a gold medal at the International Math Olympiad (IMO) this year. They will also share the specific version of this model used at the IMO with selected mathematicians and researchers. Unlike regular AI models that give answers quickly, this one spends hours thinking deeply, making it useful for serious academic research.
Google says Gemini Deep Think can help with tasks that require creative thinking, strategic planning, and step-by-step problem solving.
The model also performed very well on tough tests like Humanity’s Last Exam (HLE), scoring 34.8%, which is higher than competing AI models like xAI’s Grok 4 and OpenAI’s o3.
It can automatically use tools like code execution and Google Search, allowing it to give longer, more detailed responses than usual AI models.
Many leading AI companies, including Elon Musk’s xAI, OpenAI, and Anthropic, are now building multi-agent systems like this.
Because multi-agent AI uses more computing resources, these advanced models are usually offered only to customers with premium subscriptions.
Google plans to soon release Gemini Deep Think via an API to selected developers and businesses to see how they might use this new technology.
